Cloud microphysics—the processes governing droplets, ice, and aerosols at microscopic scales—remains a leading source of uncertainty in weather and climate predictions. These processes shape cloud structure, precipitation, and radiative feedbacks, yet they are neither resolvable in large-scale models nor directly constrained by most observing systems. Bridging the scale gap between observations, microphysical processes, and predictive models is a central challenge in atmospheric science.
This talk presents a data-driven approach to integrating satellite observations, high-fidelity simulations, and subgrid-scale modeling to better constrain and represent cloud microphysics across scales. First, I will present recent work that applies machine learning to infer vertically resolved cloud properties from passive satellite observations, combining multi-modal data sources to reconstruct properties and structure that are not directly observed. This approach provides new observation-informed constraints across regimes and helps bridge the gap between sparse microphysical measurements and global passive observations. Next, I will introduce a complementary framework for learning reduced-order representations of cloud microphysics from high-fidelity particle-based simulations. This surrogate model discovers compact, physically-consistent, and performant representations of cloud droplet dynamics, enabling efficient and interpretable representations of warm cloud processes in weather and climate models. Together, these approaches illustrate a pathway for integrating observations and simulations within a unified data-driven framework. By linking space-based observation to vertically-resolved cloud properties and reduced-order parameterization to droplet-scale details, this work aims to improve the physical realism and predictive capability of atmosphere prediction systems.